Recommendation systems represent one of the largest machine learning applications on the planet -- industry-scale recommendation models are trained with petabytes of data and serve billions of users every day. To utilize the rich user signals in the long user history, these models have been scaled up to unprecedented complexity, up to trillions of floating-point operations (TFLOPs) per example. This scale, coupled with the huge amount of training data, necessitates new storage and training algorithms to efficiently improve the quality of these complex recommendation systems. In this paper, we present a Request-Only Optimizations (ROO) training and modeling paradigm. ROO simultaneously improves the storage and training efficiency as well as the model quality of recommendation systems. We holistically approach this challenge through co-designing data, infrastructure, and model architectures. Our ROO training and modeling paradigm treats a user request as a unit of the training data. Compared with the established practice of treating a user impression as a unit, this new design directly removes redundant features in data logging, saving data storage. Second, by de-duplicating computations and communications across multiple impressions in a request, this new paradigm enables highly scaled-up neural network architectures to better capture user interest signals, such as Generative Recommenders (GRs) and other request-only friendly architectures. Our proposed ROO training and modeling paradigm has been deployed to multiple major recommendation products, each with billions of active users. ROO data format allows for increasing the training sample volume up to 2.5x. ROO training optimization yields a substantial increase in training throughput -- up to 6.7x. Combined with ROO-based neural architectures like Hierarchical Sequential Transduction Units (HSTU), ROO scales model FLOPs by 7x utilizing the same amount of training compute, enabling significant offline and online metric wins.